AIRPET: Virtual Positron Emission Tomography

J. Renner, J.J. Gómez Cadenas, S.R. Soleti · 2026

Read AIRPET: Virtual Positron Emission Tomography by J. Renner, J.J. Gómez Cadenas, S.R. Soleti (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-22059.txt.

Introduction

J. Renner, J. J. Gómez Cadenas, and S. R. Soleti build a virtual version of a medical scanner. Positron emission tomography is a powerful imaging method, but designing and evaluating new scanners is slow and costly. The work splits into detector design and simulation, image reconstruction, and image interpretation.

Their 2026 preprint presents AIRPET, a virtual PET environment that ties those stages together in software. A faithful simulator lets engineers test scanner designs before building hardware, and lets reconstruction algorithms be graded against known ground truth. That speeds the cycle of improvement for machines used in cancer diagnosis and brain research, where better images change real treatment decisions.

Catalog record

CREATORJ. Renner, J.J. Gómez Cadenas, S.R. Soleti
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE13 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-22059

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

AIRPET: Virtual Positron Emission Tomography J. Renner, J.J. Gómez Cadenas, S.R. Soleti Abstract Positron Emission Tomography (PET) is a powerful medical imaging technique, but the design and evaluation of new PET scanner technologies present significant challenges. The process is typically divided into three major stages: 1. detector design and simulation, 2. image reconstruction, and 3. image interpretation. Each of these stages requires significant expertise, making it difficult for individuals or small teams to manage all three at once. AIRPET (AI-driven Revolution in Positron Emission Tomography) is a web-based platform designed to address this challenge by integrating all phases of PET design into a single, accessible, and AI-assisted workflow. AIRPET provides an interface to large language models (LLMs) for assisted geometry creation and an interface for basic PET image reconstruction with the potential for further expansion. Here we introduce AIRPET and outline its current functionality and proposed additions. Keywords PET \cdot Geant4 \cdot GDML Code: https://github.com/jerenner/airpet 1 Introduction Positron Emission Tomography (PET) is an important technique in modern medical diagnostics, providing functional imaging of metabolic processes within the body [ 1 , 2 ] . The development of new PET scanners is a multidisciplinary endeavor involving three principle stages: 1. Detector Design and Simulation: This first stage involves setting up the PET detector model, including its geometry and materials. This is done with a simulation toolkit such as Geant4 [ 3 , 4 ] . The relevant PET physics, including electron-positron annihilation and propagation of the resulting 511 keV gamma rays through the detector geometry, must then be simulated via Monte Carlo methods for a given radiotracer distribution. The effects of detector reconstruction such as position and energy resolution must also be added. 2. Image Reconstruction: The locations of the detected gamma ray interactions, must be processed using reconstruction algorithms to obtain a 3D image of the simulated distribution. This step can be carried out using a code package such as [ 5 , 6 , 7 ] . 3. Image Interpretation: The reconstructed image must be evaluated from a medical standpoint to interpret the results of the scan. This step is performed by a medical professional. Stages (1) and (2) require a specialized skillset, creating a significant barrier to entry for researchers and engineers. To address this challenge, we have developed AIRPET, an AI-assisted platform for design and simulation of PET detectors. By integrating design, simulation, and image reconstruction into a single interface, AIRPET facilitates access to PET research and development. In addition, AIRPET aims to provide both a training ground (generation of training sets along the lines discussed above), and an AI partner, whose diagnoses can be compared with those of radiologists. The design of a modern PET scanner, such as a total-body PET (TBPET) system [ 8 ] , requires careful consideration of performance, cost, and clinical utility. As detailed in the development of the CRYSP TBPET scanner [ 9 ] , choices in scintillator materials, detector geometry, and electronics all have significant effects on the final image quality and diagnostic capabilities. The CRYSP project, for instance, explores the use of cryogenic cesium iodide crystals as a cost-effective alternative to more traditional, expensive materials, demonstrating the type of design choices that require extensive simulation and evaluation. The current PET design landscape consists of a variety of specialized software tools [ 10 ] for each stage of the process, but they require significant expertise to operate and integrate. This creates a siloed environment where experts in one domain may struggle to fully appreciate the implications of their work on the others. An integrated tool that bridges these gaps would lead to a more efficient design process. Stage (3) is performed by a medical professional. One way in which new technologies can be useful for radiologists is by providing them with an “AI partner” and a “training environment.” While in many cases training can and will occur by examining real labeled data, AI tools combined with simulations can be used to create variants and combinations of real cases, as well as new, simulated, realistic problems for diagnosis (which are verifiable, since one has the true information). 2 AIRPET: A Unified Platform for PET Design AIRPET is being developed as an open-source, web-based application that assists the user through the PET design and evaluation workflow. The project has already implemented basic functionality in detector geometry specification and allows for execution of basic simulations and PET image reconstruction. The end-to-end simulation and reconstruction workflow is as follows: 1. The user defines a geometry and simulation configuration in the web-based GUI (see Fig. 1 ). 2. The application generates the necessary files to launch the corresponding detector physics simulation. The resulting output can be analyzed directly by the user and can be useful even outside of PET physics, and/or the user can continue a PET physics analysis with step 3 below. 3. The user launches the PET reconstruction interface, which allows for computation of lines of response (LORs) and execution of the 3D PET image reconstruction from those LORs. The reconstructed 3D image array is saved and can be interactively viewed slice-by-slice in the frontend (see Fig. 2 ). Figure 1: Example of the geometry interface in AIRPET, showing the CRYSP1M [ 9 ] detector geometry and visualisation of several simulated tracks. 2.1 Detector Design AIRPET operates on a client-server model in which the frontend (HTML/Javascript) serves as the user interface for geometry manipulation and simulation preparation, while the backend (Python/Flask) manages the project state, simulation, and image reconstruction. The initial and most developed component of AIRPET is a robust toolchain focused on the creation of detector geometry. It begins with a visual editor employing three.js [ 11 ] for 3D rendering that allows for intuitive interaction with the geometry. It uses an internal JSON-based representation for saving and versioning of designs while also including support for import and export of the standard Geant4 GDML [ 12 ] format. Users can add, delete, and modify components within the scope of GDML, including defines, materials, solids (both primitive and complex booleans), logical and physical volumes, and visualization attributes. A key feature of AIRPET is the integration of an AI assistant to allow for geometry creation via text input. Currently locally-hosted open-source models (via Ollama [ 13 ] ) and cloud-based models from Google Gemini [ 14 ] are supported. A user can, for example, simply request to “create a minimal PET geometry consisting of a ring of 16 crystals with a 90 cm radius,” and the AI backend will attempt to translate this command into the necessary JSON structure to specify the geometry. Note that this feature in practice cannot currently replace a skilled human designer, and the quality of the responses depends heavily on the model employed and the user-specified prompt. While the results may not be perfect, even with current top-of-the-line models, AIRPET provides the tools to perform a significant fraction of the work with AI and then fine-tune the result manually. AIRPET also allows for CAD import via STEP files. This feature uses the pythonocc [ 15 ] library to parse STEP files, converting them into Geant4-compatible tessellated solids. 2.2 Detector Simulation AIRPET is integrated with Geant4 via a pre-compiled simulation

Read the complete text